Acute changes in DNA methylation in relation to 24 h personal air pollution exposure measurements: A panel study in four European countries

Environment International 120 (2018) 11–21

Contents lists available at ScienceDirect

Environment International journal homepage: www.elsevier.com/locate/envint

Acute changes in DNA methylation in relation to 24 h personal air pollution exposure measurements: A panel study in four European countries

T

Nahid Mostafavia, Roel Vermeulena,g, Akram Ghantousb, Gerard Hoeka, Nicole Probst-Henschc,d, Zdenko Hercegb, Sonia Taralloe, Alessio Naccaratie, Jos C.S. Kleinjansf, Medea Imbodenc,d, Ayoung Jeongc,d, David Morleyg, Andre F.S. Amaralh, Erik van Nunena, John Gulliverg, ⁎ Marc Chadeau-Hyama,g, Paolo Vineise,g, Jelle Vlaanderena, a

Division of Environmental Epidemiology, Institute for Risk Assessment Sciences, Utrecht University, 3584 CM Utrecht, the Netherlands Epigenetics Group, International Agency for Research on Cancer, Lyon, France Swiss Tropical and Public Health Institute, Basel, Switzerland d University of Basel, Basel, Switzerland e Italian Institute for Genomic Medicine (IIGM), Turin, Italy f Department of Toxicogenomics, Maastricht University, Maastricht, the Netherlands g Medical Research Council-Public Health England Centre for Environment and Health, Department of Epidemiology and Biostatistics, Imperial College London, London, United Kingdom h Population Health and Occupational Disease, National Heart and Lung Institute, Imperial College London, London, UK b c

A R T I C LE I N FO

A B S T R A C T

Handling Editor: Martí Nadal

Background: One of the potential mechanisms linking air pollution to health effects is through changes in DNAmethylation, which so far has mainly been analyzed globally or at candidate sites. Objective: We investigated the association of personal and ambient air pollution exposure measures with genome-wide DNA-methylation changes. Methods: We collected repeated 24-hour personal and ambient exposure measurements of particulate matter (PM2.5), PM2.5 absorbance, and ultrafine particles (UFP) and peripheral blood samples from a panel of 157 healthy non-smoking adults living in four European countries. We applied univariate mixed-effects models to investigate the association between air pollution and genome-wide DNA-methylation perturbations at single CpG (cytosine-guanine dinucleotide) sites and in Differentially Methylated Regions (DMRs). Subsequently, we explored the association of air pollution-induced methylation alterations with gene expression and serum immune marker levels measured in the same subjects. Results: Personal exposure to PM2.5 was associated with methylation changes at 13 CpG sites and 69 DMRs. Two of the 13 identified CpG sites (mapped to genes KNDC1 and FAM50B) were located within these DMRs. In addition, 42 DMRs were associated with personal PM2.5 absorbance exposure, 16 DMRs with personal exposure to UFP, 4 DMRs with ambient exposure to PM2.5, 16 DMRs with ambient PM2.5 absorbance exposure, and 15 DMRs with ambient UFP exposure. Correlation between methylation levels at identified CpG sites and gene expression and immune markers was generally moderate. Conclusion: This study provides evidence for an association between 24-hour exposure to air pollution and DNAmethylation at single sites and regional clusters of CpGs. Analysis of differentially methylated regions provides a promising avenue to further explore the subtle impact of environmental exposures on DNA-methylation.

Keywords: DNA-methylation Air pollution Fine particles Ultrafine particles Immune markers

⁎

Corresponding author at: Division of Environmental Epidemiology, Institute for Risk Assessment Sciences (IRAS), Utrecht University, PO Box 80178, 3508 TD Utrecht, the Netherlands. E-mail addresses: [email protected] (N. Mostafavi), [email protected] (R. Vermeulen), [email protected] (A. Ghantous), [email protected] (G. Hoek), [email protected] (N. Probst-Hensch), [email protected] (Z. Herceg), [email protected] (S. Tarallo), [email protected] (J.C.S. Kleinjans), [email protected] (M. Imboden), [email protected] (A. Jeong), [email protected] (D. Morley), [email protected] (A.F.S. Amaral), [email protected] (E. van Nunen), [email protected] (J. Gulliver), [email protected] (M. Chadeau-Hyam), [email protected] (P. Vineis), [email protected] (J. Vlaanderen). https://doi.org/10.1016/j.envint.2018.07.026 Received 23 March 2018; Received in revised form 17 July 2018; Accepted 17 July 2018 0160-4120/ © 2018 Elsevier Ltd. All rights reserved.

Environment International 120 (2018) 11–21

N. Mostafavi et al.

Table 1 Characteristics of study participants. Characteristic Session (N samples)b 1 2 Sex (N individuals) Female Male Education (N individuals) Secondary school University Age (years; median and P25–P75) BMI (kg/m2; median and P25–P75) Physical activity (MET; median and P25–P75)d Season (N samples) 1: spring (21/3–20/6) 2: summer (21/6–20/9) 3: autumn (21/9–20/12) 4: winter (21/12–20/3)

Italya

Netherlandsa

Switzerlanda

United Kingdoma

Pooledc

43 42

41 40

48 44

23 20

155 146

22 21

34 7

25 23

15 8

96 59

30 13 60 (56–63) 24.8 (22.5–26.6) 1.64 (1.5–1.7)

7 34 62 (56–68) 24.6 (22.6–27.1) 1.65 (1.5–1.7)

4 44 60 (53–68) 25.1 (21.6–26.9) 1.46 (1.4–1.6)

11 12 63 (57–65) 26.8 (24.3–29.3) 1.62 (1.5–1.8)

25 103 61 (55–66) 25 (22.5–27.9) 1.6 (1.5–1.7)

34 16 0 35

18 31 32 0

35 18 7 32

15 16 12 0

102 81 51 67

a In Italy individuals lived in Turin, in the Netherlands individuals lived in Utrecht and Amsterdam, in Switzerland individuals lived in Basel, and in the United Kingdom individuals lived in Norwich. b Personal exposure monitoring and blood collection were performed twice for each individual. Session 1 and 2 were separated by a couple of months (in average 28 days (P25–P75): (20–37) between repeated sample collection). c Our study population consisted of 157 non-smoking, healthy adults. Of which 144 individuals had two measurements and 13 individuals had one measurement. In total, 301 samples were analyzed and included in the statistical analyses. d Physical activity was measured using the accelerometer (Actigraph GT3X+) and expressed in the ‘The Metabolic Equivalent of tasks’ (MET).

1. Introduction

related to air pollution exposure (NO2) of the mother during pregnancy (Gruzieva et al., 2016). To estimate exposure to air pollution, these methylome-wide studies used either spatially resolved models that predict long-term (1 year) average levels (MESA, EPIC), or temporally resolved (2-day, 7-day, and 28-day trailing averages) estimates based on citywide monitoring data (KORA). The relation between 24-hour personal measurements of air pollution and peripheral blood methylome-wide changes is currently unknown. Our study contributes to the existing literature by providing insight into the relation between 24-hour personal measurements of particulate matter (PM) smaller than 2.5 μm (PM2.5), PM2.5 absorbance (as a proxy of black carbon and soot), and ultrafine-particulates (UFP) and genomewide changes in the peripheral blood methylome. For comparison purposes, our study also incorporates modeled long-term average estimates of the same pollutants. Associations with air pollutants were identified through genomewide univariate screening of the peripheral-blood methylome and by analyses of differentially methylated regions (DMRs). PM from different pyrogenic sources may have toxicological similarities, we, therefore, assessed the overlap between our results and a set of CpG sites previously reported to be associated with cigarette smoke. In addition, to gain a better understanding of the functional role of differential methylation at air pollutant related CpG sites, we investigated their association with expression data of closely located genes and a set of 13 immune markers measured in the same individuals.

Air pollution is now the world's largest single environmental health risk, leading to approximately 9 million deaths in 2015 alone (16% of all deaths worldwide) (Landrigan et al., 2017). The death burden relates mostly to the risk of cardiovascular and respiratory diseases (Brunekreef and Holgate, 2002; Hoek et al., 2013) and to lung cancer (Loomis et al., 2013). Oxidative damage, inflammation, and endothelial dysfunction have been suggested as potential underlying mechanisms (Demetriou et al., 2012; Pettit et al., 2012), though the identity and the role of potentially involved biological pathways are far from clear. It has been proposed that systemic effects of air pollution might be detected through assessing variation in epigenetics which can provide further mechanistic insights in air-pollution-health associations (Baccarelli et al., 2009; Herceg et al., 2017; Nawrot and Adcock, 2009). The most studied and best understood epigenetic modification is DNAmethylation, the covalent addition of a methyl group to a cytosine primarily in the context of a cytosine-guanine dinucleotide (CpG) (Bellavia et al., 2013; De Prins et al., 2013). Early studies have provided indications for an impact of air pollution on peripheral blood global methylation using repetitive elements in long (LINE-1) and short (ALU) interspersed nuclear element (Baccarelli et al., 2009; Bellavia et al., 2013; De Prins et al., 2013). Furthermore, sequence-specific analyses have yielded evidence for differential methylation in regions of genes such as inducible nitric oxide synthase (iNOS), tissue factor intercellular adhesion molecule 1, toll-like receptor 2, interferon-y, and interleukin-6 (Bind et al., 2014; Madrigano et al., 2012; Tarantini et al., 2009; Wang et al., 2016). In recent years, several cohort studies have conducted peripheral blood methylome-wide association studies of air pollution using the Illumina 450 BeadChip platform (Gruzieva et al., 2016; Panni et al., 2016; Plusquin et al., 2017). The KORA F3, F4 cohort (N = 2300 individuals), the Normative Aging Cohort (N = 657 individuals), the Multi-Ethnic Study of Atherosclerosis (MESA) (N = 1207 individuals), and the EPIC study (N = 613 individuals) identified several sets of CpG sites at which levels were associated with exposure to air pollution (Chi et al., 2016; Panni et al., 2016; Plusquin et al., 2017). In addition, a meta-analysis of multiple European and North American studies identified a set of three CpGs for which methylation levels in children were

2. Material and methods 2.1. Study population Our study, which is part of the EXPOsOMICS project (Vineis et al., 2016), was conducted in four European countries (Italy, Netherlands, Switzerland, and United Kingdom) from December 2013 to February 2015. We aimed to recruit 40 individuals from each country of whom half lived on busy roads (road with > 10 K vehicles per day; house on ground/first floor) and the other half on quiet roads (at least 100 m away from busy roads) in order to increase contrast in air pollutant levels. 157 healthy never-smoking adults between 50 and 70 years old, with no history of a pulmonary or cardiovascular disease, diabetes, or 12

Environment International 120 (2018) 11–21

N. Mostafavi et al.

each PEM session. Collected blood samples were processed, aliquoted, and transferred to −80 °C within 2 h.

other acute or chronic health conditions participated in our study (Table 1). Participants performed three personal exposure monitoring (PEM) sessions in different seasons (summer, winter, and spring/autumn) spread over one year (on average 28 days (P25–P75): (20–37) between repeated sample collection). During each PEM session, air pollution measurements and blood samples were collected for each subject. Subjects performed their own daily routine and filled in a questionnaire on food intake and time-activity. Blood samples from the first two PEM sessions were analyzed. Ethics approval for each country was obtained from local authorized review boards. All subjects signed an informed consent before participation.

2.3.1. DNA-methylation measurements Genome-wide DNA-methylation analyses were performed on peripheral blood samples (buffy coats) using the Illumina Infinium HumanMethylation450 BeadChip (HM450K), with optimized robotic pipelines (Ghantous et al., 2014; Morris and Beck, 2015). The HM450K array measures DNA-methylation at 485,512 cytosine positions across the human genome (CpG). Samples were randomized within each country. All laboratory procedures were carried out at the Epigenetics Group, International Agency for Research on Cancer (IARC, Lyon, France) according to manufacturers' protocols. CpG sites with > 10% missing methylation values were replaced by the average methylation level of that particular CpG site. After quality control exclusions, 459,333 CpGs were selected for further analyses. The percentage of methylation of a given cytosine (β-value) corresponds to the ratio of the methylated signal over the sum of the methylated and unmethylated signals. The M-value was calculated as the logit transformation of the βvalue. The M-value was used in the statistical analysis because it is less heteroscedastic and performs better in differential methylation analysis (Du et al., 2010). For each CpG, extreme values were considered as outliers if they fell out of the interquartile range (IQR) (2.5th percentile −3 × IQR, 97.5th percentile +3 × IQR) interval and were removed from further analysis. The maximum percentage of outlier samples was 3.7% in < 10% of methylation markers. DNA-methylation sites were mapped to genes based on data provided by Illumina (Bibikova et al., 2011) (for the individual probes) and on the February 2009 human reference sequence database (GRCh37) using the UCSC genome browser (Meyer et al., 2013; Pedersen et al., 2013) (both for the individual probes and the DMRs). Functional analyses of mapped genes that were identified in either the univariate analysis or DMRs to be associated with air pollution were assessed using NIH-DAVID bioinformatics resources (Huang et al., 2009).

2.2. Exposure assessment During each PEM session, participants carried a backpack containing air pollution sensors to measure 24-hour personal air pollution exposure and a belt to measure locations (GPS) and an accelerometer. Simultaneously, with the personal measurements, 24-hour air pollution was measured outdoor at the subject's home address. 2.2.1. PM2.5 For the personal measurements, 24-hour PM2.5 was sampled on a 2 μm pore size Teflon filter, using a BGI 400 pump unit and GK 2.05 SH (BGI Inc., Waltham MA, USA) cyclone. Changes in filter weight (pre and post-weighted) measured in a central lab were used to assess PM2.5 concentration. Ambient PM2.5 was collected via installing the same devices as used for personal sampling, directly outside of the subject's home in the garden or balcony. Filter reflectance was measured to assess PM2.5 absorbance concentration (soot levels) in both personal and home outdoor monitoring. Details on analytical procedures have been published previously (Eeftens et al., 2012). 2.2.2. Ultrafine particles (UFP) Ultrafine particles (UFP) were continuously monitored with a MiniDiSC (Testo AG, Lenzkirch, Germany). The MiniDiSC operated at a flow of 1000 ml/min measuring particles from 10 to 300 nm at 1second intervals. We used the term UFP to refer to the median of particle number counts from the MiniDiSC over 24-hour PEM session. Additionally, in The Netherlands and Switzerland ambient UFP was collected over each 24-hour PEM session at the subject's home facade using MiniDiSCs. Details on methods and quality control have been published previously (van Nunen et al., 2017).

2.3.2. Gene expression Total RNA was isolated from stabilized blood specimens (400 μl of whole blood and 1600 μl of RNA later) using RiboPureTM-Blood (Ambion), according to the manufacturer's instructions. RNA from 301 samples was hybridized on Agilent 8 × 60 K Whole Human Genome microarrays for mRNA. Raw data on pixel intensities were extracted using Agilent Feature Extraction Software. After quality control exclusion, a total of 23,557 probes were available for the subset (245 of 301) samples. Further details are provided in the Supplemental materials ‘2’.

2.2.3. Average exposure and long-term modeled exposure In addition to the main personal exposure metrics, reflecting exposure to air pollution in the 24 h before blood collection, we calculated the average of the 24-hour personal air pollution measurements across the three PEM sessions as a proxy for long-term average personal exposure. We also used estimated annual average ambient concentrations of air pollution for each individual using the land-use-regression (LUR) prediction models of the ESCAPE study (European study of cohorts for air pollution effects) and EXPOsOMICS (Eeftens et al., 2012; van Nunen et al., 2017) as a proxy for long-term modeled ambient exposure. Further details are provided in the Supplemental materials ‘1’.

2.3.3. Immune markers assessment A panel of 23 immune markers was measured in serum samples of all subjects using an R&D Systems (Abingdon, UK) Luminex screening assay. The panel included interleukins (IL) 1β, IL-4, IL-5, IL-6, IL-8, IL10, IL-13, IL-17, IL-25, tumor necrosis factor alpha (TNF-α), eotaxin, IL1 receptor antagonist (IL1ra), CXC chemokine ligand 10 (CXCL10), epidermal growth factor (EGF), fibroblast growth factor beta (FGF-β), granulocyte colony-stimulating factor (G-CSF), melanoma growth stimulatory activity/growth-related oncogene (GRO), chemokine (CeC motif) ligand 2 (CCL2), CeC motif chemokine 22 (CCL22), macrophage inflammatory protein-1 beta (MIP-1β), vascular endothelial growth factor (VEGF), Myeloperoxidase (MPO), and periostin. In addition, Creactive protein (CRP) was assessed using the R&D System Solid Phase ELISA. The panel of immune markers was a priori selected based on an informal review of the literature on air pollution, asthma, CVD, colon cancer, and lung cancer. This panel was applied in several studies part of the EXPOSOMICS project (Vineis et al., 2016). After quality control assessment, 13 immune markers remained for further analysis. Immune markers were log-transformed as the distributions were skewed. Further details are provided in the Supplemental material ‘3.1’.

2.2.4. Physical activity During each PEM session, physical activity was continuously measured with an accelerometer (Actigraph GT3X+, Pensacola FL, USA), which was attached to the participant's belt. We estimated energy expenditure in metabolic equivalent of task (MET) values from the raw acceleration data to express participant's physical activity over 24 h before blood collection (Ainsworth et al., 2000). 2.3. Biological sample collection Biological samples were collected in the morning (9:00–11:00) after 13

Environment International 120 (2018) 11–21

N. Mostafavi et al.

PM2.5 exposure among a reference set of CpG sites associated with “exposure to cigarette smoke” (Joehanes et al., 2016) using the gene set enrichment methodology (Mostafavi et al., 2017; Subramanian et al., 2005). A p-value < 0.05 was used as the statistical cut-off for enrichment. A further description of these reference sets is provided in the Supplemental materials ‘5.1’.

2.4. Data analysis 2.4.1. Univariate mixed-effects model We used linear mixed effect models setting a random intercept (i) for each subject, capturing the correlation among measurements within the same subject (N = 2 observation per subject), (ii) for each microtiter plate (N = 37) and (iii) for the position of the sample on the chip (N = 12), capturing technically-induced variation potentially biasing DNA-methylation measurements. To control for potential confounding, models were adjusted for sex, age, body mass index (BMI) (kg/m2), education (primary, secondary school, and university), country (Italy, Netherlands, Switzerland, and United Kingdom), season and physical activity expressed in MET-values. In addition, temperature and relative humidity as proxies of meteorological conditions were included in all models. A number of sensitivity analyses were conducted. First, we ran a minimally adjusted model (only age, sex, and BMI included as covariates) and a model in which country was excluded from the primary set of covariates. To explore the possible influence of country, we ran our original model for each country separately and on the pooled set of data leaving data from one country out one at the time. We explored the stability of the methylation levels over time by using the random effects part of the mixed model to estimate intra-class correlation coefficients (ICC; the proportion of the total variance in the CpG sites that was due to variation between subjects, and not due to variation between plates, position on the plates, or error). We also assessed the association between DNA-methylation and long-term average personal exposures as well as long-term modeled exposure to air pollution in two separate models with the same set of covariates as were considered in the original model. Missing values in covariate and exposure data were imputed using multivariate imputation by chained equations (MICE) in R (Buuren and GroothuisOudshoorn, 2011). No covariate had > 4% missing data (except for MET; 13%). In exposure data, 6.3% of personal PM2.5 and 5.6% of ambient PM2.5 values were missing. Since the proportion of missing data was low, only one imputation iteration was used in association analysis. White blood cell composition was estimated based on the DNAmethylation data, using the Houseman algorithm (Houseman et al., 2012). Potential confounding by white blood cell composition was evaluated by assessing the association between white blood cell composition and the air pollutants using the linear mixed-models as described above. Multiple testing adjustments on the resulting p-values were performed using the false discovery rate method of Benjamini-Hochberg (BH-FDR) (Benjamini and Hochberg, 1995) at the level of 0.05. Univariate mixed-effects models and imputation were conducted using R version 3.0.2 (packages: lme4, mice (De Boeck et al., 2011; Buuren and Groothuis-Oudshoorn, 2011)).

2.4.4. Impact of air pollution on immune markers We assessed the association between air pollution measurements and immune markers (n = 13) using univariate mixed-effects models. We included random intercepts for each subject which captures the correlation among measurements within the same subject and a random intercept for each microtiter plate (N = 9) which captures the nuisance variation generated in the assessment of the immune markers in the model. The same set of covariates as we defined in our primary univariate methylation models were used to control for potential confounding variables. Moreover, the same sensitivity analysis as we described in the methylation analysis was performed to assess the stability of our findings to variation in the confounder models. A further description of these analyses is provided in the Supplemental materials ‘3.2’. 2.4.5. Functional readout of the impact of differential methylation at the identified CpG sites We assessed interrelationships between methylation levels at significant CpG sites identified in the methylation analysis (both univariate and DMRs analysis for each air pollutants separately; 415 CpG sites), expression of mapped genes (n = 58 transcripts), and the immune markers (n = 13) incorporated in this study using the XMWAS package (Uppal et al., 2017). We set the minimum pairwise correlation threshold to 0.4 and used a canonical Partial Least Squares regression (10 components) to generate pair-wise similarity matrices (González et al., 2012). We used a clustering approach to address the paired measurements in this analysis. Connections with a correlation < 0.4 were not shown in the figure. The correlation structure between the three platforms was displayed using the Cytoscape software environment (Shannon et al., 2003). 3. Results Baseline characteristics of the study participants are presented in Table 1. Our study population consisted of 157 non-smoking, healthy adults for whom 301 blood samples and air pollutant measurements were collected across two sampling campaigns (144 participants with two measurements and 13 participants with one measurement). About 61% of participants were female and > 65% of participants had a university or college degree. The population had a median (P25–P75) age of 61 (55–66) years. In Fig. 1, we show the distribution of 24-h average personal and ambient air pollution measurements by country. Highest air pollutant levels were measured in Italy, followed by The Netherlands. Lower levels were measured in the United Kingdom and in Switzerland (Fig. 1). We observed a relatively high correlation between measured ambient and personal air pollution concentration measurements, especially for PM2.5 absorbance (r = 0.74). Correlations between short-term measured air pollution concentrations and modeled long-term air pollution concentrations were low. Correlations between UFP and the other air pollutants were weak to moderate (r = 0.18; Fig. S1). We did not observe any significant association between the air pollutants and estimated white blood cell type composition (Table S2); therefore, adjustment for white blood cell type composition was not used in further analyses.

2.4.2. Differentially methylated regions (DMRs) We identified DMRs using the comb-p package in Python (version 2.7.13) (Pedersen et al., 2012). p-Values for each probe generated by the univariate linear mixed-effect models (ordered by chromosomal location) were used as input. We set a window size of 300 bases and restricted DMRs to those that included at least 2 probes. A Stouffer–Liptak–Kechris (SLK) and Šidák corrected p-value < 0.05 was used as the statistical cut-off for significance. This resulted in regions of adjacent probes with low p-values that stand up to genome-wide correction. Regions were annotated to nearest gene and CpG island from human genome version “hg19” using the CruzDB software package (Meyer et al., 2013; Pedersen et al., 2013). A further description of this analysis is provided in Supplemental material ‘4’.

3.1. Univariate analyses 2.4.3. Overlap with CpG sites associated with cigarette smoking We assessed the distribution of CpG sites associated with personal

We identified 13 CpG sites (Mapped to 14 genes) at which 14

Environment International 120 (2018) 11–21

N. Mostafavi et al.

Fig. 1. Box plots of air pollution concentrations by country. Each panel shows one air pollution marker; personal PM2.5; μg/m3, personal PM2.5 absorbance; m−1, personal UFP; particles/cm3, ambient PM2.5; μg/m3, ambient PM2.5 absorbance; m−1, and ambient UFP; particles/cm3. Each box represents one country (Italy, Netherlands, Switzerland, and the United Kingdom). Horizontal lines correspond to medians, and boxes to the 25th–75th percentiles; whiskers extend to data within the interquartile range times 1.5, and data beyond this are plotted as dots.

Fig. 2. Manhattan plot showing fixed-effect p-values of association between personal PM2.5 measurements and DNA-methylation for fully adjusted model. Each dot corresponds to the CpG methylation site. Horizontal lines correspond to adjusted p-value (FDR) at 0.05 level. 13 CpG sites above the line were considered statistically significant. Table 2 CpG sites significantly (at FDR 5%) associated with personal measurements of PM2.5. Probe ID (CpG)

βa

p-Value

FDR

Chromosome

Location

Mapped gene (region)b

Relation to CpG island

Nearby genesc

cg26692818 cg02556634 cg03873392 cg07669973 cg02508204 cg19850855 cg20673255 cg23468453 cg26559703 cg01905633 cg03408122 cg05404940 cg20693615

0.23 0.06 −0.13 −0.12 0.20 0.20 0.14 0.23 0.24 0.06 0.13 −0.11 0.23

4.1E−08 1.0E−07 2.4E−07 3.0E−07 6.8E−07 7.8E−07 4.7E−07 7.1E−07 6.1E−07 6.3E−07 1.3E−06 1.2E−06 1.2E−06

0.02 0.02 0.03 0.03 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04

1 4 16 10 11 5 19 10 10 6 22 11 3

87,108,066 6,449,000 10,801,987 135,029,365 39,367,436 93,333,655 5,787,465 73,906,532 26,853,625 3,849,391 50,903,314 1,446,911 50,234,688

CLCA3P (gene body) PPP2R2C (gene body)

OpenSea OpenSea OpenSea Island OpenSea OpenSea N_Shore OpenSea N_Shelf Island Island OpenSea S_Shelf

CLCA3P; LOC105378828 PPP2R2C TEKT5 KNDC1 LINC01493 FAM172A DUS3L ASCC1 APBB1IP FAM50B SBF1 BRSK2 GNAT1

a b c

KNDC1 (gene body) FAM172A (gene body) DUS3L (gene body) ASCC1 (gene body) APBB1IP (gene body) FAM50B (TSS1500) SBF1 (gene body) BRSK2 (gene body) GNAT1 (3′UTR)

The coefficient estimate from a model with M values as the outcome. Annotations provided by Illumina (Bibikova et al., 2011). Annotation to nearest gene from human genome version hg19 with CruzDB package (Meyer et al., 2013; Pedersen et al., 2013).

15

Environment International 120 (2018) 11–21

N. Mostafavi et al.

Fig. 3. (A) Association between personal PM2.5 and methylation of 13 identified CpG sites for a different set of confounder adjustment including Model 1 (full model) adjusted for sex, age, BMI, education, season, physical activity, temperature, relative humidity, and country. Model 2 adjusted for sex, age, BMI, and country. Model 3 adjusted for all variables of Model 1 except for country. (B) Association between personal PM2.5 and methylation of 13 identified CpG sites in pooled population and stratified by country. (C) Association between personal PM2.5 and methylation of 13 identified CpG sites in pooled population and after leaving out countries one-byone. In Italy individuals lived in Turin, in the Netherlands individuals lived in Utrecht and Amsterdam, in Switzerland lived in Basel, and in the United Kingdom lived in Norwich.

the 13 CpG sites was generally low (median 0.13), indicating considerable within individual variation. No clear correlation patterns in methylation levels were observed between the 13 CpG sites (Fig. S2). No significant associations were identified with any of the other

methylation levels were significantly associated (FDR < 0.05) with personal exposure to PM2.5 (Fig. 2). For ten sites methylation increased with increasing exposure to PM2.5, while for the remaining sites methylation decreased (Table 2). Between subject intraclass correlation for 16

Environment International 120 (2018) 11–21

N. Mostafavi et al.

3.4. Impact of air pollution on immune markers

personal or ambient air pollution measurements (Table S3). However, coefficients for the 13 CpG sites associated with personal PM2.5 in the models for the other air pollution markers were largely comparable reaching often nominal statistical significance at the p < 0.05 level, especially for personal PM2.5 absorbance and ambient PM2.5 (Table S4). Sensitivity analyses indicated that the 13 associations found for PM2.5 were robust to the set of confounders considered (Fig. 3A). Stratified analyses using data from each country separately yielded consistent effect size estimates in sign and magnitude (Fig. 3B). However, estimated strength of association was generally weakened upon stratification and revealed some heterogeneity across countries. Overall, of the 13 CpG sites identified in the pooled analysis, all were significantly associated to personal PM2.5 in the Dutch population, 9 in the Basel population, 4 in the Turin population, and one in the Norwich population. The weakening of the strength of the associations can partially be attributed to the smaller sample size due to stratification. To account for this we ran a further sensitivity analysis excluding data from each country one-by-one. Results confirmed consistent effect size estimates with some evidence of stronger effects in The Netherlands for some of the CpGs (Fig. 3C). Nominally significant (p < 0.05) effect estimates from a model that included the long-term average personal PM2.5 exposure were correlated (r = 0.77; Fig. S3) to the nominally significant effect estimates from our main model based on 24-h personal measurements for PM2.5, though none of the associations identified in this analysis were genomewide significant. No correlation was observed with nominally significant effect estimates from a model that included long-term modeled exposure to PM2.5 (r = 0.02; Fig. S3) and no genome-wide significant associations were identified.

We identified a significant association between ambient exposure to PM2.5 absorbance and serum concentration of CCL22 (β = 0.11; FDR = 0.03; Fig. S5A/B). The other pollutants yielded no significant association with any of the immune markers. Some indications of an association (FDR < 0.2) were observed between personal exposure measurements of UFP and serum concentration of two additional immune markers [CXCL10 (β = 0.16; FDR = 0.15) and G.CSF (β = 0.32; FDR = 0.15)], which are nevertheless noteworthy as comparable effects were observed for these markers in an independent population (PISCINA) that was part of the EXPOsOMICS project (Fig. S5A). The observed associations in that study were in the same direction as in our main study and significant (at FDR < 0.2) for G.CSF (0.25; FDR = 0.17) but not for CXCL10 (0.34; FDR = 0.28). The observed associations for CCL22, G.CSF, and CXCL10 were robust in a series of sensitivity analyses (Fig. S6). Further description of these results is provided in the Supplemental materials ‘3.3’. 3.5. Functional readout of the impact of differential methylation at the identified CpG sites Fig. 4 shows the interrelations between CpG sites significantly associated to personal PM2.5 (both in DMR (69 CpGs) and univariate analysis (13 CpGs)), expression of mapped genes, and the immune markers based on a similarity matrix generated using PLSR (González et al., 2012). Though modest in absolute correlation (absolute range of correlation was between 0.4 and 0.73), Fig. 4 illustrates a high degree of interrelatedness between the selected set of markers: 240 markers including 177 CpG sites (annotated to 56 unique genes), 55 expression of mapped genes (annotated to 41 unique genes), and 8 immune markers. We identified two clusters of considerably correlated markers and a very small weakly interrelated third cluster, which were primarily driven by the direction of the correlation. Cluster one and two contained 134 and 100 markers (nodes), respectively. Further information about clusters and interrelatedness is provided in Excel-file Table E4.

3.2. Differentially methylated regions (DMRs) We identified 162 DMRs (SLK-Sidak p-value < 0.05) of which 69 DMRs (containing 74 genes and 404 CpG sites) were significantly associated with personal exposure to PM2.5 (Table 3), 42 DMRs (42 genes and 266 CpG sites) with personal exposure to PM2.5 absorbance, 16 DMRs (21 genes and 109 CpG sites) with personal exposure to UFP, 4 DMRs (4 genes and 12 CpG sites) with ambient exposure to PM2.5, 16 DMRs (19 genes and 142 CpG sites) with ambient PM2.5 absorbance exposure, and 15 DMRs (20 genes and 125 CpG sites) with ambient exposure to UFP (Table S5 and Supplementary excel-file Tables E1, E2). Overlap of annotated genes across significant DMRs was observed between personal PM2.5 and personal PM2.5 absorbance (n = 11), and between personal PM2.5 absorbance and ambient PM2.5 absorbance (n = 10) while for other combinations overlap was less pronounced (Table S5). Two of the significant CpGs identified in our univariate analysis (Mapped to KNDC1 and FAM50B) were located within the DMRs that were associated with PM2.5.

4. Discussion Our study provides evidence that measurements of personal exposure to particulate air pollution, collected in a panel of healthy participants from four European countries, were associated with DNAmethylation changes in specific regions of the genome. Correlation between methylation at these CpG sites and a set of immune markers measured in peripheral blood was generally moderate, as was the correlation with the expression of genes mapped to the identified CpG sites. Several immune markers were independently associated with some of the air pollution measurements. Our results do not replicate findings from previous studies of short-term air pollution and blood DNA-methylation (Bellavia et al., 2013; Panni et al., 2016; Wang et al., 2018), though no consistent patterns have been observed in these studies. As we observed relatively low ICCs for the CpG sites that were associated to the PM2.5 measurements, our study points towards (partly reversible) short-term variation of DNA methylation in peripheral blood in response to daily variation in PM2.5 levels. Due to the ubiquitous presence of air pollution in our environment, studying the temporal behavior of the DNA methylome in response to exposure to air pollution in observational studies is a challenge. A handful of experimental studies among both animals and humans do indicate a short-term response to exposure to air pollution. In one experimental study, demethylation of the interleukin-2 gene of lymph node T cells was observed in vivo 20 min after antigen presentation (Bruniquel and Schwartz, 2003). Furthermore, in a bioassay conducted by (Ding et al., 2017) methylation of LINE-1, iNOS, p16CDKN2A, and APC, in both lung tissue as well as in peripheral blood, changed after four hours of exposure to PM2.5. (Bellavia et al., 2013) observed global

3.3. Overlap with CpG sites associated with cigarette smoking Cigarette smoke is a source of particulate exposure with some similar characteristics to ambient air pollution (but contributing to a much higher exposure). To assess the overlap between our results and a set of CpG sites previously reported to be associated to air pollution (Joehanes et al., 2016), we conducted an approach comparable to gene set enrichment (Subramanian et al., 2005). Hypermethylated CpG sites that were associated with cigarette smoking were significantly enriched (p-value < 0.01; Fig. S4A) among the most strongly hypermethylated CpG sites in our analysis for PM2.5. A similar enrichment was observed for smoking-related hypomethylated CpG sites in our analysis (p < 0.01; Fig. S4B). Further description of these results is provided in the Supplemental materials ‘5.2’.

17

Environment International 120 (2018) 11–21

N. Mostafavi et al.

Table 3 DMRs associated with personal measurements of PM2.5 at an SLK-Sidak p value of < 0.05. DMR

Chromosome

Start

End

No. of probes

SLK-Sidak p value

Mapped genea

Relation to CpG island

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69

1 1 1 1 1 1 1 1 10 10 10 11 11 11 11 11 12 12 12 12 12 12 12 15 15 16 16 16 16 17 18 19 19 19 19 19 19 19 19 2 2 2 2 2 2 2 2 20 22 3 3 4 4 4 4 5 5 5 6 6 6 6 6 7 7 8 8 8 9

26,231,347 167,408,509 240,656,217 153,599,479 92,414,221 244,952,269 154,474,344 153,234,037 135,029,294 14,051,636 70,321,554 1,413,145 1,456,891 1,463,541 19,736,150 1,474,588 95,945,120 33,205,702 4,918,848 71,552,188 46,319,990 47,219,626 6,649,733 83,240,550 57,510,284 85,551,478 84,691,498 89,408,076 67,686,832 33,759,484 74,799,250 45,448,959 2,888,943 55,660,514 57,742,112 23,941,207 11,649,462 45,975,694 38,281,047 239,983,744 207,506,480 38,892,846 20,870,812 64,868,515 242,763,794 236,506,413 20,870,087 26,190,328 38,092,643 39,321,710 116,163,694 74,847,646 7,657,070 174,429,263 46,995,203 157,079,312 140,719,008 1,962,311 31,938,984 3,849,190 30,038,791 529,098 126,080,132 158,905,317 56,515,510 599,963 82,644,373 26,722,496 136,149,908

26,231,585 167,409,199 240,656,668 153,600,157 92,414,911 244,952,395 154,474,801 153,234,388 135,029,462 14,052,029 70,321,960 1,413,316 1,457,347 1,463,663 19,736,334 1,474,842 95,945,385 33,205,863 4,919,231 71,552,570 46,320,111 47,220,093 6,649,995 83,240,792 57,510,576 85,551,749 84,691,851 89,408,568 67,687,120 33,760,250 74,799,573 45,449,302 2,889,257 55,660,626 57,742,445 23,941,768 11,649,702 45,976,196 38,281,560 239,984,106 207,507,164 38,893,252 20,871,402 64,868,711 242,763,983 236,506,614 20,870,363 26,190,355 38,093,208 39,322,104 116,164,243 74,848,017 7,657,709 174,429,542 46,995,744 157,079,521 140,719,303 1,962,555 31,939,547 3,849,703 30,039,901 529,342 126,080,724 158,905,537 56,516,256 600,556 82,644,769 26,722,966 136,150,033

2 7 5 8 8 3 5 4 3 6 6 3 4 4 5 3 4 3 5 4 4 11 4 7 3 3 3 6 3 10 4 5 3 5 9 7 5 5 5 4 9 6 6 2 2 2 3 3 11 3 6 8 5 6 7 5 4 3 12 21 39 3 4 3 11 6 8 5 3

9.1E−03 5.6E−03 4.4E−03 1.8E−02 3.1E−03 3.2E−02 2.7E−02 1.3E−02 4.5E−02 9.6E−03 2.1E−03 2.2E−02 7.0E−03 5.8E−04 1.6E−03 3.4E−02 9.2E−03 1.8E−02 2.3E−02 7.3E−03 3.1E−02 1.5E−02 9.3E−03 2.5E−02 1.0E−02 2.1E−02 2.8E−02 3.0E−02 2.0E−03 7.9E−06 1.1E−03 4.7E−02 3.0E−03 2.3E−02 1.1E−06 6.2E−04 2.2E−02 4.2E−02 9.4E−04 5.6E−03 3.0E−02 4.8E−04 6.2E−07 1.9E−03 8.7E−03 2.3E−02 1.1E−04 1.9E−02 4.7E−03 3.6E−02 6.1E−04 1.2E−04 6.7E−03 4.0E−02 1.9E−03 2.1E−02 3.1E−03 9.1E−04 5.5E−04 7.0E−04 3.2E−11 1.5E−02 3.9E−04 1.9E−02 4.7E−02 2.8E−06 3.3E−02 2.1E−02 3.1E−03

STMN1 CD247 MIR1273E;GREM2 S100A13 BRDT COX20 SHE;TDRD10 LOR KNDC1 FRMD4A TET1 BRSK2 BRSK2 BRSK2 LOC100126784;NAV2 BRSK2 USP44 PKP2 KCNA6 TSPAN8 SCAF11 SLC38A4 IFFO1 CPEB1 TCF12 GSE1 KLHL36 ANKRD11 CARMIL2 SLFN12 MBP APOC4-APOC2 ZNF556 TNNT1 AURKC ZNF681 CNN1 FOSB ZNF573 HDAC4 FAM237A GALM GDF7 SERTAD2 NEU4 AGAP1 GDF7 MIR663AHG TRIOBP CX3CR1 LSAMP PF4 SORCS2 SCRG1 GABRA4 SOX30 PCDHGA2;PCDHGA1 CTD-2194D22.4 DXO;STK19 FAM50B RNF39 EXOC2 HEY2 VIPR2 LOC650226 ERICH1 CHMP4C ADRA1A ABO

Shore Island Island

a

Island Island Island Shore Shore Shore Shore Shore Shore Island

Island

Island Island Island Island Island Island Island Island Island Island Island Island Shore Island Island Island

Island Shore Island Island Island Shore Island Island Island Island Island Island Island Island Island

DMRs were annotated to nearest gene from human genome version hg19 with CruzDB package (Meyer et al., 2013; Pedersen et al., 2013).

18

Environment International 120 (2018) 11–21

N. Mostafavi et al.

Methylaon Gene expression Immune markers Fig. 4. Illustration of the interrelations between the three omics platform: Methylation (CpG sites significantly associated with personal PM2.5 (both in DMR (69 CpGs) and univariate analysis (13 CpGs)); Green rectangle), immune markers (Blue rectangle) and gene expression (expression of mapped genes; Red rectangle). Three rings show different clusters (Cluster1 contained 134 omics markers and cluster 2 and 3, 100 and 6 omics markers, respectively. The connecting lines show association (correlation) between different markers (with absolutes range between 0.4, 0.73) Positive in Red; negative in Blue). The color intensity and the thickness of the lines both indicate the strength of the association (thicker lines indicate a correlation higher than 0.6). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

DNA demethylation after 130 min of exposure to fine concentrated ambient particles (Bellavia et al., 2013). (Chen et al., 2016) observed significant reduction of both global and site-specific methylation after 48 h of exposure to PM2.5 (Chen et al., 2016). (Jiang et al., 2014) reported significant changes in DNA methylation 6 h after exposure to diesel exhaust (Jiang et al., 2014). In addition to experimental evidence, several observational studies identified significant associations between short-term changes in methylation and exposure to air pollution. (Panni et al., 2016). observed demethylation to be associated with average air pollution levels in the two days before blood sampling (though a stronger signal was observed with longer trailing averages (Panni et al., 2016), while (Wang et al., 2018) observed demethylation within 12 h of exposure to PM2.5 (Wang et al., 2018). Combined, results from our and other studies suggest that the impact of air pollution on the blood methylome is likely subtle and requires high-quality exposure assessment and considerable statistical power to be studied. Strengths of our study include the application of a common study protocol and standardized operating procedures applied across four European countries. DNA-methylation, gene expression, and immune markers were each analyzed in a single laboratory and blinded duplicate samples were included for quality assurance. State-of-the-art personal exposure measurements of air pollution were conducted to reduce the impact of measurement error and we repeated measurements within a time period of a year to cover seasonal variation. High-quality exposure assessment is resource intensive. Our study was, therefore, necessarily modest in the number of individuals that could be included in the study, which is a limitation. Another limitation of our study was the fact that we were unable to identify an independent dataset in which we could replicate the methylation findings of this study. However, we did assess the robustness of the identified signals within our dataset as much as possible.

In addition to univariate regression analysis, we applied two alternative approaches (gene set enrichment and DMR) that maximized the information we could derive from our study. Both approaches overcome statistical limitations in the assessment of perturbations by combining information from multiple CpG sites rather than assessing sites one by one. Enrichment of smoking-related CpG sites among the CpG sites most strongly associated with air pollution points towards a shared biologic pathway of the effects of cigarette smoke and air pollution on the epigenome. This is of interest due to overlap between health outcomes that have been related to tobacco smoking and air pollution (Wang et al., 2015). We recently discovered a similar overlap in an analysis of another air pollutant (NOx) and the peripheral blood transcriptome (Mostafavi et al., 2017). We focused on DMRs in our analysis because CpG sites mostly function in groups to regulate gene expression, rather than independently (Breton et al., 2017). Similar to gene set enrichment, our DMR analysis combined information from multiple CpG sites rather than assessing sites one by one and therefore also increased the statistical power to detect subtle perturbations in methylation patterns, leading to a much richer signal in our DMR analysis than in our univariate regression. Two CpG sites that were independently significantly associated to PM2.5 were located within a DMR, which increased the credibility of these two associations. Acknowledging that the 450 K array chip was designed to interrogate the methylation status of CpG sites in proximity to genes, a potential approach to gain insight into the biological role of DMRs and single CpG sites associated with air pollution involves mapping them to nearby genes and conducting functional enrichment analysis on these genes (Wright et al., 2016). Functional analysis of the identified CpG sites at the mapped genes using the NIH-DAVID bioinformatics resources (Huang et al., 2009) did not provide any insight into potentially affected biologic pathways, likely due to the limited number of CpG sites found in this study. Further functional interpretation of the identified CpG sites requires moving towards regulatory enrichment 19

Environment International 120 (2018) 11–21

N. Mostafavi et al.

between short-term air pollutants and systemic inflammation in human study populations were inconsistent (Hassanvand et al., 2017; Larsson et al., 2013; Steenhof et al., 2014).

analysis, incorporating not only mapping of nearby genes, but also acknowledging trans-acting effects of DNA-methylation on gene expression (Wright et al., 2016). In this study, we observe some evidence for such complex interactions between DNA-methylation, gene expression, and immune markers by showing the correlations between these markers. The existence of three correlation clusters could point towards shared biological function, though further analyses are required to confirm this interpretation. 24-hour personal exposure measurements of PM2.5 were more strongly associated with DNA-methylation than long-term modeled exposure estimates of PM2.5. Plusquin et al., using a much larger study among healthy adults, reported a similar absence of significant association of DNA-methylation in response to long-term PM2.5, NOx, NO2 (Plusquin et al., 2017). These observations could be an indication that acute changes in methylation due to exposure to air pollution identified in this study did not reflect a long-term association between air pollution and DNA-methylation. However, long-term modeled exposure estimates could also be much more affected by measurement error which would have reduced the statistical power to detect significant associations. 24-h personal exposure measurements of PM2.5 were also more strongly associated with DNA-methylation than ambient measurements of PM2.5 and measurements of the other air pollutants. Measurements of ambient PM2.5 and PM2.5 absorbance were likely more strongly affected by measurement error. The weaker effect for personal PM2.5 absorbance is surprising as PM2.5 absorbance has been reported to be a very good marker for exposure to traffic-related particles (Cyrys et al., 2003). Direct comparison of the results for PM2.5 with those for personal UFP measurements is complex as these pollutants not only differ in particle size but also in the way they were assessed (mass weight versus counts). Differences in observed associations might, therefore, be due to measurement error but might also have a biological interpretation related to particle size. Further work is required to clarify this issue. The optimal way to summarize continuous UFP measurements in an aggregated metric is still an ongoing discussion (van Nunen et al., 2017). Exposure to UFP appears to be primarily determined by peaks and therefore summarizing exposure into a mean or median might not optimally reflect the impact of 24-hour exposure to UFP on the human system (van Nunen et al., 2017). The collected continuous UFP measurements provide opportunities for further explorations of sensitive time windows and the contribution of outdoor versus indoor sources to the total UFP exposure received (van Nunen et al., 2017). In this study, we assessed exposure in the 24 h before blood draw. Limited information is available on the time-scale on which environmental factors are expected to have an impact on changes in DNAmethylation. Therefore the time-period over which exposure should be assessed to maximize the power to assess potential associations is not known. In a study by (Wang et al., 2016, 2018), personal monitoring of PM2.5 was conducted 3 days before scheduled blood draw using a microPEM device. Based on this data different time lags were calculated, of which the 0–24 h time lag was most closely related to differential methylation. This result provides some support for the time window included in our study. Even though we did not observe any associations between exposure to air pollution and estimated cell type composition, we repeated the main analysis for PM2.5 correcting for the estimated cell type composition by including seven cell types (“Monocytes”, “B”, “CD4T”, “NK”, “CD8T”, “Eosinophils”, “Neutrophils”). Results did not change substantially (Table S6) and were strongly correlated (r = 0.92) with the effect estimates from our original model. Our study also provides evidence for perturbation of serum concentration of three immune markers (CCL22, CXCL10, and G.CSF) in association with air pollution (CCL22 with ambient PM2.5 absorbance; CXCL10 and G.CSF with personal UFP). So far, these immune markers have not been reported in association with air pollution markers in other studies. However, previously published reports on the association

5. Conclusion In conclusion, this study provide evidence for an association between 24-hour exposure to air pollution, in particular measurements of PM2.5, and DNA-methylation both at single CpG sites and DMRs. Analysis of DMRs provides a promising avenue to further explore the subtle impact of environmental exposures on DNA-methylation. Acknowledgements The research leading to these results has received funding from the European Community's Seventh Framework Program (FP7/2007e2011) under grant agreement number: 308610 (EXPOsOMICS). This work also was supported by the Swiss National Science Foundation, SNFSAPALDIA (grants number 33CS30-148470/1). We thank Dr. Manolis Kogevinas for providing the opportunity to replicate our immune marker findings in the PISCINA dataset. We greatly acknowledge all those who are responsible for data collection and management in the EXPOsOMICS study. Appendix A. Supplementary data Supplementary data to this article can be found online at https:// doi.org/10.1016/j.envint.2018.07.026. References Ainsworth, B.E., Haskell, W.L., Whitt, M.C., Irwin, M.L., Swartz, A.M., Strath, S.J., ... Leon, A.S., 2000. Compendium of physical activities: an update of activity codes and MET intensities. Med. Sci. Sports Exerc. 32 (9 Suppl), S498–S504(Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/10993420). Baccarelli, A., Wright, R.O., Bollati, V., Tarantini, L., Litonjua, A.A., Suh, H.H., ... Schwartz, J., 2009. Rapid DNA methylation changes after exposure to traffic particles. Am. J. Respir. Crit. Care Med. 179 (7), 572–578. https://doi.org/10.1164/rccm. 200807-1097OC. Bellavia, A., Urch, B., Speck, M., Brook, R.D., Scott, J.A., Albetti, B., ... Baccarelli, A.A., 2013. DNA hypomethylation, ambient particulate matter, and increased blood pressure: findings from controlled human exposure experiments. J. Am. Heart Assoc. 2 (3), 1–10. https://doi.org/10.1161/JAHA.113.000212. Benjamini, Y., Hochberg, Y., 1995. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J. R. Stat. Soc. Ser. B 57, 289–300. Retrieved from. http://www.jstor.org/stable/2346101. Bibikova, M., Barnes, B., Tsan, C., Ho, V., Klotzle, B., Le, J.M., ... Shen, R., 2011. High density DNA methylation array with single CpG site resolution. Genomics 98 (4), 288–295. https://doi.org/10.1016/j.ygeno.2011.07.007. Bind, M.-A., Lepeule, J., Zanobetti, A., Gasparrini, A., Baccarelli, A., Coull, B.A., ... Schwartz, J., 2014. Air pollution and gene-specific methylation in the Normative Aging Study: association, effect modification, and mediation analysis. Epigenetics 9 (3), 448–458. https://doi.org/10.4161/epi.27584. Breton, C.V., Marsit, C.J., Faustman, E., Nadeau, K., Goodrich, J.M., Dolinoy, D.C., ... Murphy, S.K., 2017. Small-magnitude effect sizes in epigenetic end points are important in children's environmental health studies: the children's environmental health and disease prevention research center's epigenetics working group. Environ. Health Perspect. 125 (4), 511–526. https://doi.org/10.1289/EHP595. Brunekreef, B., Holgate, S.T., 2002. Air pollution and health. Lancet 360 (9341), 1233–1242. https://doi.org/10.1016/S0140-6736(02)11274-8. Bruniquel, D., Schwartz, R.H., 2003. Selective, stable demethylation of the interleukin-2 gene enhances transcription by an active process. Nat. Immunol. https://doi.org/10. 1038/ni887. Buuren, S., Groothuis-Oudshoorn, K., 2011. Mice: Multivariate imputation by chained equations in R. J. Stat. Softw. 45 (3), 67. Retrieved from. http://doc.utwente.nl/ 78938/. Chen, R., Meng, X., Zhao, A., Wang, C., Yang, C., Li, H., ... Kan, H., 2016. DNA hypomethylation and its mediation in the effects of fine particulate air pollution on cardiovascular biomarkers: a randomized crossover trial. Environ. Int. https://doi.org/ 10.1016/j.envint.2016.06.026. Chi, G.C., Liu, Y., MacDonald, J.W., Barr, R.G., Donohue, K.M., Hensley, M.D., ... Kaufman, J.D., 2016. Long-term outdoor air pollution and DNA methylation in circulating monocytes: results from the Multi-Ethnic Study of Atherosclerosis (MESA). Environ. Health 15 (1), 119. https://doi.org/10.1186/s12940-016-0202-4. Cyrys, J., Heinrich, J., Hoek, G., Meliefste, K., Lewné, M., Gehring, U., ... Brunekreef, B., 2003. Comparison between different traffic-related particle indicators: elemental

20

Environment International 120 (2018) 11–21

N. Mostafavi et al.

W.J., 2013. The UCSC Genome Browser database: extensions and updates 2013. Nucleic Acids Res. 41 (Database issue), D64–D69. https://doi.org/10.1093/nar/ gks1048. Morris, T.J., Beck, S., 2015. Analysis pipelines and packages for Infinium Human Methylation450 Bead Chip (450 k) data. Methods 72 (C), 3–8. https://doi.org/10. 1016/j.ymeth.2014.08.011. Mostafavi, N., Vlaanderen, J., Portengen, L., Chadeau-Hyam, M., Modig, L., Palli, D., ... Vermeulen, R., 2017. Associations between genome-wide gene expression and ambient nitrogen oxides. Epidemiology 28 (3), 320–328. https://doi.org/10.1097/EDE. 0000000000000628. Nawrot, T.S., Adcock, I., 2009. The detrimental health effects of traffic-related air pollution. Am. J. Respir. Crit. Care Med. 179 (7), 523–524. https://doi.org/10.1164/ rccm.200812-1900ED. van Nunen, E., Vermeulen, R., Tsai, M.-Y., Probst-Hensch, N., Ineichen, A., Davey, M.E., ... Hoek, G., 2017. Land use regression models for ultrafine particles in six European areas. Environ. Sci. Technol. https://doi.org/10.1021/acs.est.6b05920. (acs.est. 6b05920). Panni, T., Mehta, A.J., Schwartz, J.D., Baccarelli, A.A., Just, A.C., Wolf, K., ... Peters, A., 2016. Genome-wide analysis of DNA methylation and fine particulate matter air pollution in three study populations: KORA F3, KORA F4, and the normative aging study. Environ. Health Perspect. 124 (7), 983–990. https://doi.org/10.1289/ehp. 1509966. Pedersen, B.S., Schwartz, D.A., Yang, I.V., Kechris, K.J., 2012. Comb-p: software for combining, analyzing, grouping and correcting spatially correlated P-values. Bioinformatics 28 (22), 2986–2988. https://doi.org/10.1093/bioinformatics/bts545. Pedersen, B.S., Yang, I.V., De, S., 2013. CruzDB: software for annotation of genomic intervals with UCSC genome-browser database. Bioinformatics (Oxford, England) 29 (23), 3003–3006. https://doi.org/10.1093/bioinformatics/btt534. Pettit, A.P., Brooks, A., Laumbach, R., Fiedler, N., Wang, Q., Strickland, P.O., ... Kipen, H.M., 2012. Alteration of peripheral blood monocyte gene expression in humans following diesel exhaust inhalation. Inhal. Toxicol. 24 (3), 172–181. https://doi.org/ 10.3109/08958378.2012.654856. Plusquin, M., Guida, F., Polidoro, S., Vermeulen, R., Raaschou-Nielsen, O., Campanella, G., ... Chadeau-Hyam, M., 2017. DNA methylation and exposure to ambient air pollution in two prospective cohorts. Environ. Int. 108, 127–136. https://doi.org/10. 1016/j.envint.2017.08.006. Shannon, P., Markiel, A., Ozier, O., Baliga, N.S., Wang, J.T., Ramage, D., ... Ideker, T., 2003. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 13 (11), 2498–2504. https://doi.org/10.1101/gr. 1239303. Steenhof, M., Janssen, N.A.H., Strak, M., Hoek, G., Gosens, I., Mudway, I.S., ... Brunekreef, B., 2014. Air pollution exposure affects circulating white blood cell counts in healthy subjects: the role of particle composition, oxidative potential and gaseous pollutants – the RAPTES project. Inhal. Toxicol. 26 (3), 141–165. https://doi. org/10.3109/08958378.2013.861884. Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S., Ebert, B.L., Gillette, M.A., ... Mesirov, J.P., 2005. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. U. S. A. 102 (43), 15545–15550. https://doi.org/10.1073/pnas.0506580102. Tarantini, L., Bonzini, M., Apostoli, P., Pegoraro, V., Bollati, V., Marinelli, B., ... Baccarelli, A., 2009. Effects of particulate matter on genomic DNA methylation content and iNOS promoter methylation. Environ. Health Perspect. 117 (2), 217–222. https://doi.org/10.1289/ehp.11898. Uppal, K., Go, Y.-M., Jones, D.P., 2017. xMWAS: R package for data integration and network analysis. In: xMWAS: An R Package for Data-driven Integration and Differential Net-work Analysis, https://doi.org/10.1101/122432. Vineis, P., Chadeau-Hyam, M., Gmuender, H., Gulliver, J., Herceg, Z., Kleinjans, J., ... Wild, C.P., 2016. The exposome in practice: design of the EXPOsOMICS project. Int. J. Hyg. Environ. Health. https://doi.org/10.1016/j.ijheh.2016.08.001. Wang, T.W., Vermeulen, R.C.H., Hu, W., Liu, G., Xiao, X., Alekseyev, Y., ... Lan, Q., 2015. Gene-expression profiling of buccal epithelium among non-smoking women exposed to household air pollution from smoky coal. Carcinogenesis 36 (12), 1494–1501. https://doi.org/10.1093/carcin/bgv150. Wang, C., Chen, R., Cai, J., Shi, J., Yang, C., Tse, L.A., ... Kan, H., 2016. Personal exposure to fine particulate matter and blood pressure: a role of angiotensin converting enzyme and its DNA methylation. Environ. Int. 94, 661–666. https://doi.org/10.1016/j. envint.2016.07.001. Wang, C., Chen, R., Shi, M., Cai, J., Shi, J., Yang, C., ... Weinberg, C.R., 2018. Possible mediation by methylation in acute inflammation following personal exposure to fine particulate air pollution. Am. J. Epidemiol. https://doi.org/10.1093/aje/kwx277. Wright, M.L., Dozmorov, M.G., Wolen, A.R., Jackson-Cook, C., Starkweather, A.R., Lyon, D.E., York, T.P., 2016. Establishing an analytic pipeline for genome-wide DNA methylation. Clin. Epigenetics 8, 45. https://doi.org/10.1186/s13148-016-0212-7.

carbon (EC), PM2.5 mass, and absorbance. J. Expo. Anal. Environ. Epidemiol. 13, 134–143. https://doi.org/10.1038/sj.jea.7500262. De Boeck, P., Bakker, M., Zwitser, R., Nivard, M., Hofman, A., Tuerlinckx, F., Partchev, I., 2011. The estimation of item response models with the lmer function from the lme4 package in R. J. Stat. Softw. 39 (12), 1–28. https://doi.org/10.18637/jss.v039.i12. De Prins, S., Koppen, G., Jacobs, G., Dons, E., Van de Mieroop, E., Nelen, V., ... Schoeters, G., 2013. Influence of ambient air pollution on global DNA methylation in healthy adults: a seasonal follow-up. Environ. Int. 59, 418–424. https://doi.org/10.1016/j. envint.2013.07.007. Demetriou, C.A., Raaschou-Nielsen, O., Loft, S., Møller, P., Vermeulen, R., Palli, D., ... Vineis, P., 2012. Biomarkers of ambient air pollution and lung cancer: a systematic review. Occup. Environ. Med. 69 (9), 619–627. https://doi.org/10.1136/oemed2011-100566. Ding, R., Jin, Y., Liu, X., Ye, H., Zhu, Z., Zhang, Y., ... Xu, Y., 2017. Dose- and time-effect responses of DNA methylation and histone H3K9 acetylation changes induced by traffic-related air pollution. Sci. Rep. https://doi.org/10.1038/srep43737. Du, P., Zhang, X., Huang, C.-C., Jafari, N., Kibbe, W.A., Hou, L., Lin, S.M., 2010. Comparison of beta-value and M-value methods for quantifying methylation levels by microarray analysis. BMC Bioinf. 11 (1), 587. https://doi.org/10.1186/1471-210511-587. Eeftens, M., Beelen, R., De Hoogh, K., Bellander, T., Cesaroni, G., Cirach, M., ... Hoek, G., 2012. Development of land use regression models for PM2.5, PM2.5 absorbance, PM10 and PM coarse in 20 European study areas; results of the ESCAPE project. Environ. Sci. Technol. 46, 11195–11205. Ghantous, A., Saffery, R., Cros, M.-P., Ponsonby, A.-L., Hirschfeld, S., Kasten, C., ... Hernandez-Vargas, H., 2014. Optimized DNA extraction from neonatal dried blood spots: application in methylome profiling. BMC Biotechnol. 14 (1), 60. https://doi. org/10.1186/1472-6750-14-60. González, I., Cao, K.-A.L., Davis, M.J., Déjean, S., 2012. Visualising associations between paired “omics” data sets. BioData Mining 5 (1), 19. https://doi.org/10.1186/17560381-5-19. Gruzieva, O., Xu, C.-J., Breton, C.V., Annesi-Maesano, I., Antó, J.M., Auffray, C., ... Melén, E., 2016. Epigenome-wide meta-analysis of methylation in children related to prenatal NO2 air pollution exposure. Environ. Health Perspect. 104 (July), 104–110. https://doi.org/10.1289/EHP36. Hassanvand, M.S., Naddafi, K., Kashani, H., Faridi, S., Kunzli, N., Nabizadeh, R., ... Yunesian, M., 2017. Short-term Effects of Particle Size Fractions on Circulating Biomarkers of Inflammation in a Panel of Elderly Subjects and Healthy Young Adults. https://doi.org/10.1016/j.envpol.2017.02.005. Herceg, Z., Ghantous, A., Wild, C.P., Sklias, A., Casati, L., Duthie, S.J., ... HernandezVargas, H., 2017. Roadmap for investigating epigenome deregulation and environmental origins of cancer. Int. J. Cancer. https://doi.org/10.1002/ijc.31014. Hoek, G., Krishnan, R.M., Beelen, R., Peters, A., Ostro, B., Brunekreef, B., Kaufman, J.D., 2013. Long-term air pollution exposure and cardio-respiratory mortality: a review. Environ. Health 12 (1), 43. https://doi.org/10.1186/1476-069X-12-43. Houseman, E.A., Accomando, W.P., Koestler, D.C., Christensen, B.C., Marsit, C.J., Nelson, H.H., ... Kelsey, K.T., 2012. DNA methylation arrays as surrogate measures of cell mixture distribution. BMC Bioinf. 13 (1), 86. https://doi.org/10.1186/1471-210513-86. Huang, D.W., Sherman, B.T., Lempicki, R.A., 2009. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat. Protoc. 4 (1), 44–57. https://doi.org/10.1038/nprot.2008.211. Jiang, R., Jones, M.J., Sava, F., Kobor, M.S., Carlsten, C., 2014. Short-term diesel exhaust inhalation in a controlled human crossover study is associated with changes in DNA methylation of circulating mononuclear cells in asthmatics. Part. Fibre Toxicol. https://doi.org/10.1186/s12989-014-0071-3. Joehanes, R., Just, A.C., Marioni, R.E., Pilling, L.C., Reynolds, L.M., Mandaviya, P.R., ... London, S.J., 2016. Epigenetic signatures of cigarette smoking clinical perspective. Circ. Cardiovasc. Genet. 9 (5), 436–447. https://doi.org/10.1161/CIRCGENETICS. 116.001506. Landrigan, P.J., Fuller, R., Acosta, N.J.R., Adeyi, O., Arnold, R., Basu, N.N., ... Zhong, M., 2017. The Lancet Commission on pollution and health. Lancet 391 (10119), 462–512. https://doi.org/10.1016/S0140-6736(17)32345-0. Larsson, N., Brown, J., Stenfors, N., Wilson, S., Mudway, I.S., Pourazar, J., Behndig, A.F., 2013. Airway inflammatory responses to diesel exhaust in allergic rhinitis. Inhal. Toxicol. 25 (3), 160–167. https://doi.org/10.3109/08958378.2013.765932. Loomis, D., Grosse, Y., Lauby-Secretan, B., El Ghissassi, F., Bouvard, V., BenbrahimTallaa, L., ... Straif, K., 2013. The carcinogenicity of outdoor air pollution. Lancet Oncol. 14 (13), 1262–1263. https://doi.org/10.1016/S1470-2045(13)70487-X. Madrigano, J., Baccarelli, A., Mittleman, M.A., Sparrow, D., Spiro, A., Vokonas, P.S., ... Schwartz, J., 2012. Air pollution and DNA methylation: interaction by psychological factors in the VA normative aging study. Am. J. Epidemiol. 176 (3), 224–232. https://doi.org/10.1093/aje/kwr523. Meyer, L.R., Zweig, A.S., Hinrichs, A.S., Karolchik, D., Kuhn, R.M., Wong, M., ... Kent,

21